Machine setup and best practices for Tensorflow-GPU

Question

I would like to run Tensorflow operations on a Nvidia 960 graphics card. Previously, I tried installing CUDA, cuDNN, and Tensorflow-GPU, but it quickly became cumbersome and I found many of the installation issues I ran into were Windows-related.

While I would like to avoid primarily opinion-based answers, what is a solid, compatible, and up to date "baseline" version and type for the following necessaries?

1) Linux distribution

2) CUDA version

3) cuDNN version

4) Python version and manager (Conda? Standard pip?)

What are some installations best-practices for easy use?

If this is better suited for Cross-Validated or another SE, please let me know.

Answer 1

Here is what I ended up using (for anyone stumbling on this post).

1) Ubuntu 16.04

2) CUDA 8.0

3) cuDNN 6.0

4) Anaconda